hpDJ: An Automated DJ with Floorshow Feedback

Abstract

Many radio stations and nightclubs employ Disk-Jockeys (DJs) to provide a continuous uninterrupted stream or mix of dance music, built from a sequence of individual song-tracks. In the last decade, commercial pre-recorded compilation CDs of DJ mixes have become a growth market. DJs exercise skill in deciding an appropriate sequence of tracks and in mixing 'seamlessly' from one track to the next. Online access to large-scale archives of digitized music via automated music information retrieval systems offers users the possibility of discovering many songs they like, but the majority of consumers are unlikely to want to learn the DJ skills of sequencing and mixing. This paper describes hpDJ, an automatic method by which compilations of dance-music can be sequenced and seamlessly mixed by computer, with minimal user involvement. The user may specify a selection of tracks, and may give a qualitative indication of the type of mix required. The resultant mix can be presented as a continuous single digital audio file, whether for burning to CD, or for play-out from a personal playback device such as an iPod, or for play-out to rooms full of dancers in a nightclub. Results from an early version of this system have been tested on an audience of patrons in a London nightclub, with very favourable results. Subsequent to that experiment, we designed technologies which allow the hpDJ system to monitor the responses of crowds of dancers/listeners, so that hpDJ can dynamically react to those responses from the crowd. The initial intention was that hpDJ would monitor the crowds reaction to the song-track currently being played, and use that response to guide its selection of subsequent song-tracks tracks in the mix. In that version, its assumed that all the song-tracks existed in some archive or library of pre-recorded files. However, once reliable crowd-monitoring technology is available, it becomes possible to use the crowd-response data to dynamically remix existing song-tracks (i.e, alter the track in some way, tailoring it to the response of the crowd) and even to dynamically compose new song-tracks suited to that crowd. Thus, the music played by hpDJ to any particular crowd of listeners on any particular night becomes a direct function of that particular crowds particular responses on that particular night. On a different night, the same crowd of people might react in a different way, leading hpDJ to create different music. Thus, the music composed and played by hpDJ could be viewed as an emergent property of the dynamic interaction between the computer system and the crowd, and the crowd could then be viewed as having collectively collaborated on composing the music that was played on that night. This en masse collective composition raises some interesting legal issues regarding the ownership of the composition (i.e.: who, exactly, is the author of the work?), but revenue-generating businesses can nevertheless plausibly be built from such technologies.